Artificial neural networks for temporal processing applied to prediction of electric energy in small hydroelectric power stations

Abstract

The purpose of this work is to present a computational prediction of temporal series through artificial neural networks (ANN) with temporal features based on short-term memory structures and episodic long-term memory. The connectionist prediction is applied to a Brazilian small hydroelectric power station, with generation capacity of 15 MWh, because conventional prediction statistical techniques show inadequacy in relation to noise, acquisition fails, and need for generalization, when applied to this model. Departing from the proposed system, it is intended also to develop, in the future, a non-linear complex system, employing ANNs, with the inclusion of new variables in the decision process, in addition to the episodic memory model, which is considered computationally feasible with the current available resources.